Le 09/05/2016 17:34, clipka a écrit :
> Am 09.05.2016 um 07:48 schrieb Le_Forgeron:
>> Le 09/05/2016 à 02:43, clipka a écrit :
>>
>>  and am wondering what the uv_mapping
>>> rules might be (I think, ideally they should match those of the "cone"
>>> primitive).
>>>
>>
>> Ah, Ah... very funny. ROTFL and far more. Really the best joke ever.
> 
> I concede I didn't have a closer look at the cone implementation; just
> intended to make sure as early as possible in the design process that
> the UV mapping isn't inconsistent with stuff we already have.
> 
> Obviously, yeah, since we don't seem to have UV mapping for the cone in
> the first place, that turned out to be a moot point.
> 
> Please don't just grab something randomly out of a bag though. What I
> see in your sample image looks like a ratio of 1:3:1 for the three
> components of the object; since this is arbitrary anyway, I would
> suggest a ratio of 1:2:1 instead. Not only does this slightly simplify
> the maths the user will have to do, and also makes those operations
> mathematically more stable (a floating-point division by any power of 2
> is lossless; a divison by 5 never is) -- much more importantly, it is a
> far better fit for standard image sizes commonly used for UV mapping
> (which tend to be powers of 2; even where image sizes are multiples of
> 5, they'll typically also be multiples of 4.)

I have no experience with image used for uv mapping, so I will trust you on that.
Initially (a code nobody have seen), I used a 10% for each end, but changed it
for a 20%, based on current VAT.
Updating for 25% is not a problem, it just seems a waste of pixels when
most of them get squeeze near the centre of the disc.

Let's listen to other's opinions first.

> 
> We should also consider whether we want the UV mapping of the main body
> to just use a cylindrical mapping (as seems to be the case in your
> sample image, though that might of course be an optical illusion), or
> whether we want the mapping to be equidistant along what I presume to be
> the V coordinate (essentially boiling down to a variation of toroidal
> mapping).
> 

Yep, cylindrical mapping so far for the torus part. I'm just dreaming of the
rocket on the cover (and inside) the "Destination moon" of Tintin (by Hergé).
And it was the easiest to code.

If we want the toroidal mapping, the code would need to store the uv with the
intersection as I have a feeling that the computation could be painful to revert.
Yet, it would be a waste of cycles when uv is not used.

Post a reply to this message